Feature extraction from images of endoscopic large intestine

In this paper, we propose feature extraction methods from two types of images of endoscopic large intestine taken by a colonoscopy for diagnosis of colon cancer. Today, there are two observation methods. One is staining surface of large intestine. The other is colonoscopy using Narrow Band Imaging (NBI) system, a new feature of endoscope. We describe extraction methods of features for each observation method so that the features may be used to estimate colon cancer staging from an observed image. Pit pattern is a texture that appears on the surface of stained intestine and they are categorized and used for diagnosis. Thus, we extract pits from an endoscope image to analyze patterns. First, color edge of the image is extracted, then watershed segmentation is applied. In the result, pits are roughly extracted. NBI system can observe vasucular structure under the surface of large intestine. The vascular structure can be used to estimate cancer staging. A vascular area is roughly extracted by adaptive binarization, then the fine shape of vascular area is extracted by the level set method